If you want to learn about batch inferencing
Article compares performance of large data center cards. Tells us for larger batches , inference speed of individual requests goes down, but overall performance of the entire batch goes up. In other words more tokens per second for the batch, but lower TPS for each request.